Bayesian Correlation is a robust similarity measure for single cell RNA-seq data

Abstract

<jats:title>Abstract</jats:title><jats:p><jats:bold>Assessing similarity</jats:bold> is highly important for bioinformatics algorithms to determine correlations between biological information. A common problem is that similarity can appear by chance, particularly for low expressed entities. This is especially relevant in single cell RNA-seq (scRNA-seq) data because read counts are much lower compared to bulk RNA-seq.</jats:p><jats:p>Recently, a <jats:bold>Bayesian correlation</jats:bold> scheme, that assigns low similarity to genes that have low confidence expression estimates, has been proposed to assess similarity for bulk RNA-seq. Our goal is to extend the properties of the Bayesian correlation in scRNA-seq data by considering 3 ways to compute similarity. First, we compute the similarity of pairs of genes over all cells. Second, we identify specific cell populations and compute the correlation in those populations. Third, we compute the similarity of pairs of genes over all clusters, by considering the total mRNA expression.</jats:p><jats:p>We demonstrate that Bayesian correlations are more reproducible than Pearson correlations. Compared to Pearson correlations, Bayesian correlations have a smaller dependence on the number of input cells. We show that the Bayesian correlation algorithm assigns high similarity values to genes with a biological relevance in a specific population.</jats:p><jats:p>We conclude that Bayesian correlation is a robust similarity measure in scRNA-seq data.</jats:p>

Toward Modeling Context-Specific EMT Regulatory Networks Using Temporal Single Cell RNA-Seq Data

“…We generated a series of gene networks for each condition by varying the cutoff values of MI. From our initial exploration, we found that activating links were favored in the network construction, probably because of the nature of scRNA-seq data (Sanchez-Taltavull et al, 2019). To select different numbers of activating and inhibitory links, we varied the MI cutoffs for positive and negative interactions.…”

scLink: Inferring Sparse Gene Co-expression Networks from Single-cell Expression Data

“…Accurate and robust estimation of the strength of gene co-expression relationships is the key to reliable inference of gene co-expression networks. Since single-cell gene expression data contain excess zero counts and relatively inaccurate low counts due to both technical and biological variability [35,29], the conventional Pearson or Spearman's correlation coefficients are often not reliable for single-cell gene expression data, especially for genes whose expression values are highly sparse ( Figure 1) [36,20]. In our previous work, we proposed a statistical method, scImpute, to address the excess zeros in scRNA-seq data [29].…”

“…However, these two measures cannot provide a robust estimation of gene co-expression given the sparse scRNA-seq data with substantial technical noises and biological heterogeneity [20,21].…”